Kilo Code

Kilo Code is an open-source AI coding agent — Apache-2.0 licensed — that runs across VS Code, JetBrains IDEs, and the CLI. It's a fork in the Cline / Roo Code lineage of VS Code extensions that have spent the last 18 months iterating on the agentic-coding pattern: read the codebase, plan the change, edit across files, run tests, iterate. Kilo combines features from Cline and Roo with its own additions around custom modes, Kilo Gateway (500+ models in one API surface), and cloud agents that run 24/7 outside your laptop.

The headline pricing math, since 'cline pricing' and 'best free coding ai agent' show up as recovering queries. Kilo Auto is the free tier — a curated default model with no credit card required and no hard token cap, generous enough to evaluate the product on real work. The BYOK path is the right one for serious daily use: connect your Anthropic, OpenAI, or other model API key to Kilo Gateway and pay providers directly. A solo developer on BYOK with Claude Sonnet typically lands at $30–$80/month in API costs depending on how heavy the workflow is. There's no agent platform fee on top of that — Kilo itself is free open source.

Kilo Code vs Cline vs Roo Code — the comparison anyone evaluating this lineage ends up making. **Cline** is the upstream project, the largest community, the most mature documentation. The default choice if you're new to agentic VS Code extensions. **Roo Code** is a Cline fork with opinionated mode-based workflows — Code mode, Architect mode, Ask mode — that change how the agent behaves. The right pick if you want stronger guard-rails around what the agent does. **Kilo Code** combines features from both and adds JetBrains support, the Kilo Gateway model router, and cloud agents. The right pick if you need JetBrains, want broader model access in one place, or want async agents triggered from Slack. All three are Apache 2.0; switching between them is low cost.

Kilo Code vs Cursor — the more interesting comparison for builders not committed to VS Code extensions. Cursor is a VS Code fork with the AI baked in; Kilo is an extension that runs on standard VS Code (and JetBrains, and CLI). Cursor's product polish and editor-level UX are ahead — that's been earning $2B in annualised revenue. Kilo's edge is portability: same agent across VS Code, IntelliJ, and the terminal, no editor lock-in. If you're a JetBrains shop or a multi-IDE team, Kilo is the only serious agentic option. If you're a VS Code shop happy paying $20/month, Cursor's UX is hard to beat.

Kilo Code vs Claude Code — a real question for builders who already pay for Claude. Claude Code is included in Claude Pro ($20/month) and is the polished, official Anthropic product. It runs Claude only. Kilo Code on BYOK with an Anthropic key runs the same models and costs the same per-token (you're paying Anthropic directly either way) but lets you switch to OpenAI, Gemini, or local models mid-session. The trade-off: Claude Code is more polished and has the deepest integration with the Claude model family; Kilo gives you model flexibility and JetBrains support. If you only use Claude and only use VS Code or terminal, stick with Claude Code. If you want options, Kilo.

What Kilo Code is genuinely good at: agentic refactors in JetBrains (the only mature option for IntelliJ / WebStorm / PyCharm users), cost-conscious solo development on BYOK, async agent workflows triggered from Slack or Discord, teams that want to A/B test models on real coding tasks without re-tooling. The Kilo Gateway abstraction makes 'try the same prompt across Claude Sonnet 4.6 and GPT-5' a one-click action, which is genuinely useful when picking a default model for your team.

Where Kilo falls short: documentation depth lags Cline's (Cline has been around longer with more contributors), the JetBrains plugin while real is still catching up to the VS Code experience in feature parity, and the 500+ model menu can be decision-overload for new users who'd be better served by a curated default. Cloud agents are a newer feature and less battle-tested than CI/CD-triggered agents for high-stakes automated work.